Long-Term Trends of Instability and Associated Parameters Over The

1 Long-term trends of instability and associated parameters

2 over the Indian region obtained using radiosonde network 3 Rohit Chakraborty, Madineni Venkat Ratnam* and Shaik Ghouse Basha

4 National Atmospheric Research Laboratory, India. 5 Correspondence to: M. Venkat Ratnam ([email protected]) 6

7 Abstract 8 Long-term trends of the parameters related to convection and instability obtained from 27 radiosonde 9 stations across 6 sub-divisions over the Indian region during the period 1980-2016 is presented. A total of 16 10 parcel and instability parameters along with moisture content, wind shear, and thunderstorm and rainfall 11 frequencies have been utilized for this purpose. Robust fit regression analysis is employed on the regional 12 average time series to calculate the long-term trends on both seasonal and yearly basis. The Level of Free 13 Convection (LFC) and Equilibrium Level (EL) height is found to ascend significantly in all Indian sub- 14 divisions. Consequently, the coastal regions (particularly the western coasts) experience increasing in Severe 15 Thunderstorm (TSS) and Severe Rainfall Frequencies (SRF) in the pre-monsoon while the inland regions 16 (especially central India) experience an increase in Ordinary Thunderstorm (TSO) and Weak Rain Frequency 17 (WRF) during the monsoon and post-monsoon. The 16-20 year periodicity is found to dominate the long-term 18 trends significantly compared to other periodicities and the increase in TSS, and Convective Available Potential 19 Energy (CAPE) is found more severe after the year 1999. The enhancement in moisture transport and associated 20 cooling at 100 hPa along with dispersion of boundary layer pollutants is found to be the main cause for the 21 increase in CAPE which leads to more convective severity in the coastal regions. However, in inland regions 22 moisture-laden winds are absent and the presence of strong capping effect of pollutants on instability in the 23 lower troposphere has resulted in more Convective Inhibition Energy (CINE). Hence, TSO and weak rainfall 24 occurrences have increased particularly in these regions. 25 Key words: Instability, Convection, Long-term trends, Radiosonde 26 27 1. Introduction

28 Intense convective phenomena are a common climatic feature in the Indian tropical region which occurs 29 during the pre-monsoon to post-monsoon seasons (April–October) (Ananthakrishnan, 1977) and they are 30 generally accompanied by intense thunderstorms, lightning, wind gusts with heavy rainfall. Hence, they are 31 known to induce immense socio-economic hazards including loss of life and property. Several reports have 32 shown an increase in the climatic extreme occurrence and intensity of these phenomena throughout the world 33 (Webster et al., 2005; Emanuel, 2006). In this connection, the traditional surface-based parcel theory has been 34 utilized to understand convective processes using atmospheric soundings as it calculates the atmospheric 35 instabilities and other parameters at various heights (Huntrieser et al., 1997; Santhi et al., 2014; Nelli et al, 36 2018a).

37 Considering the importance of studying the long-term trends in climatic extremes, a series of research 38 attempts have been orchestrated world-wide in the last two decades. Using multiple tropical stations and re- 39 analysis data, Gettleman et al. (2002) and Riemann-Campe et al. (2009) have shown that Convective Available 40 Potential Energy (CAPE) has been increasing very strongly with a growth rate of ~20% per decade during the 41 period 1958-1997 due to increase in surface heating and moisture. Gensini and Mote (2015) projected a 236 % 42 increase in severe thunderstorm frequency from 1980-1990 to 2080-2090 over the eastern United States (US). 43 Further, Brooks (2013) used various combinations of CAPE and Vertical Wind Shear (VWSH) products and 44 results hinted towards a probable increase in the severe thunderstorms over the US. It was also observed that the 45 effect of increasing CAPE is more dominant on convective severity than in case of decreasing shear. On the 46 other hand, studies by Prein et al. (2017) showed that a recent increase of temperature has led to a rise of 47 moisture ingress and consequently the frequency and severity of extreme precipitation events associated with 48 intense convection have shown a steep rise everywhere in the world. At the same time, an increase in 49 thunderstorm severity and instability has also been reported by many attempts over the Asian region (Wang et 50 al., 2011; Saha et al., 2017). 51 Over the Indian region, Manohar et al. (1999) studied the latitudinal variation and distribution of 52 thunderstorm frequency and CAPE over 78 Indian stations during 1970-1980 and they postulated that the 53 ambient temperature at 100 hPa pressure level has a strong relationship with it. Dhaka et al. (2010) utilized 54 radiosonde observations during 1958-1997 and obtained very prominent anti-correlations on both yearly and 55 seasonal basis between convection strength (CAPE) and upper troposphere temperatures at 100 hPa (T100). 56 Later, Murugavel et al. (2012) studied the long term trends of CAPE from 32 radiosonde stations during 1984- 57 2008 and revealed an alarming growth in monsoon CAPE over India with a slope of 38 J/kg/year. However, 58 they additionally stated that the low-level moisture and solar cycle can have additional impact on the increasing 59 CAPE. Recently from reanalysis datasets, Chakraborty et al. (2017a) and Saha et al. (2017) reported that lower 60 lower tropospheric instability is reducing over few Indian stations after 1980 due to increasing levels of 61 pollution. Apart from that, some studies have also attempted to correlate convective severity with boundary 62 layer phenomena, surface fluxes, solar effect and precipitation; (Murthy and Sivaramakrishnan, 2006; 63 Allappattu and Kunnikrishnan, 2009; Xie et al., 2011, Nelli et al. 2018b). 64 Previous studies over India have shown the distribution of CAPE only whereas other parameters like 65 Convective Inhibition Energy (CINE), Mixed Layer CAPE (MLC), Lifted Index (LI), Total Totals Index (TTI), 66 and Precipitable Water Vapor (PWV) are also important as they explain how the atmospheric instability and 67 moisture changes at various levels of the atmosphere. In addition, the influence of climatic oscillation (Quasi- 68 Biennial Oscillation (QBO), El-Nino Southern Oscillation (ENSO) and Solar Cycle) on the seasonal and annual 69 variation of convective parameters was also not studied in detail. Therefore in the present study, long-term 70 variation of parcel parameters (Lifted Condensation Level (LCL), Level of Free Convection (LFC), Equilibrium 71 Level (EL), CAPE and CINE), with instability (LI, Vertical Totals Index (VT)), moisture (PWV, and PWV at 72 low levels (PWL)), thunderstorm and rainfall severity frequencies (Thunderstorm-Severe (TSS), Thunderstorm- 73 Ordinary (TSO), Weak Rain Fall (WRF) and Strong Rain Fall (SRF)) followed by Temperature at 100hPa 74 (T100) and Wind Shear (WSH) is investigated using 27 radiosonde stations along with gridded rainfall data over 75 India. This article is structured as follows: Section 2 describes the datasets and methodology adopted for the 76 present study. Section 3 presents the long-term analysis of parcel and instability parameter over Chennai

77 (13.08oN, 80.27oE) and 6 sub-divisions of the Indian subcontinent on both annual and seasonal basis, followed 78 by the periodicity and split trend analysis. Finally, a discussion on the results and conclusions is appended in 79 Section 4 and 5, respectively. 80 81 2. Dataset and Methodology

82 Radiosonde observations from 27 stations over the Indian region from 1980-2016 are obtained from 83 Integrated Global Radiosonde Archives (https://www1.ncdc.noaa.gov/pub/data/igra/derived/derived-por/). 84 These datasets provide daily temperature and humidity profiles from 1538 stations around the world in fixed 85 pressure levels after doing quality checks (Durre et al., 2006; Ferreira et al. 2018). These studies have concluded 86 that the radiosonde data quality from IGRA has faced certain problems from time to time, but such cases are not 87 so prominent over the Indian region, especially after the year 1980. It is mainly because of the higher accuracy 88 and reliability of this in-situ measurement technique that these datasets are widely used worldwide nowadays for 89 calibrating other continuous profiler instruments (Chakraborty and Maitra, 2016). In accordance with data 90 availability and reliability, only 27 stations have been considered out of 37 IGRA Indian radiosonde stations 91 thereby providing descent data availability for carrying out this study. When an in depth investigation is done on 92 the data continuity by plotting the temperature and humidity profiles for all days, a set of gaps in datasets were 93 noticed. Most of the utilized stations have intermittent data gaps of 2-7 days in certain months only but, on the 94 whole, except only a very few cases, the duration of these individual data gaps are mostly limited to less than 1 95 month. However, these small data gaps are not expected to provide any significant impact on the long-term 96 seasonal or annual average variations of 37 years x 12 months span. 97 In addition to the data availability, homogeneity also acts as a common concern before using the data. 98 However, such issues should not be considered serious as all three types of homogeneity issues namely: volume, 99 instrument type and quality have been addressed before commencing the study. First, about 5000 radiosonde 100 profiles are available in majority of IGRA stations which are uniformly distributed among all years and seasons 101 (except monsoon); hence it provides a decent data volume for investigation of yearly trends. Secondly, the data 102 of all Indian IGRA stations come from a single type of IM-MK3 radiosondes which has not undergone any 103 change in radiosonde accuracies in the last years and so this addresses the instrument type related issue. Finally, 104 regarding data quality, a set of 7 quality checks are performed by IGRA before accepting the data which should 105 remove any unreliable observations before being used in the study. These 7 quality checks also include 106 repetition check which rejects any possible case of humidity sensor saturation errors during rainy conditions 107 especially in monsoon. Thus it can be inferred that the obtained climatic trends of instability from IGRA is 108 expected to be far more reliable compared to other data sources. 109 These 27 stations have been divided into six homogenous regions as defined by the India 110 Meteorological Department (IMD) (Rao, 1976) which are: Central India (CI), East Coast (EC), North East (NE), 111 North West (NW), Peninsula India (PI) and the West Coasts (WC) as shown in Fig. 1. Further, for simplicity, 112 these regions have again been combined into three major categories namely: coastal regions (EC and WC), 113 Inland (CI and PI) and others (NE and NW). After retrieving the profiles some more additional internal quality 114 checks are performed before using the data for every station. First, the balloon burst height has to be minimum 115 of 15 km to be selected for analysis. Second, any gaps of temperature and humidity in important pressure levels 116 such as 850, 700, 500, 300, 200 and 100 hPa will produce difficulty in calculation of atmospheric instability;

117 hence those profiles were rejected in quality control tests for all stations. Again, out of the available radiosonde 118 profiles, some profiles have displayed absurd variations of temperature and humidity at various heights and 119 hence they are discarded. After completion of these quality checks it was thought that atmospheric instability 120 shows prominent diurnal variation, datasets of only one time slot can be taken for analysis. Finally as datasets at 121 00Z are consistently much more in number than at 12Z, hence analysis has been actually done with 00Z 122 datasets. A complete detail about the final dataset used for every station is indicated in Table A1. It may be 123 noted here that the volume of observations are found to distributed almost homogenously throughout the 124 measurement period and a detailed year wise breakup of radiosonde launches utilized are not shown to maintain 125 the focus of the work. For calculation of the instability parameters, the temperature and humidity profiles were 126 transformed from the standard pressure levels using cubic spline interpolation at every 100 m height bins. Piece- 127 wise linear/ quadratic/ cubic spline interpolation schemes are employed instead of linear interpolation in 128 temperature and humidity retrievals in this study as the former techniques can more faithfully regenerate the 129 nonlinearities in boundary layer variations of meteorological parameters according to recent studies by 130 Chakraborty et al. (2016).. After this, a similar surface-based parcel method is utilized for estimating the parcel 131 and instability parameters (LCL, LFC, EL, CAPE, MLC, CINE) as already described by Chakraborty et al. 132 (2018). A small detail about the physical significance of these parameters is now given in the Supplementary 133 Section. For thunderstorm genesis, moisture growth and wind shear are extremely important, therefore we 134 calculated the total amount of water vapor (PWV) and that up to 700 hPa level (PWL) along with the horizontal 135 wind shear between surface and 6 km altitude. In addition to these, we have used temperature at the 100 hPa 136 pressure level as it is found to strongly influence the convective strengths over the Indian region (Manohar et al., 137 1999; Dhaka et al., 2010). 138 Along with these parameters, the long-term impact of instability on the convection has also been 139 studied from thunderstorm and rain frequencies. Daily measurements of surface wind speeds is obtained for all 140 the radiosonde observations at 00Z using the Wyoming Website (weather.uwyo.edu/upperair/sounding.html). 141 The thunderstorm frequencies are calculated on yearly basis based on the criterion given by IMD 142 (http://imd.gov.in/section/nhac/termglossary.pdf). According to this criterion, if the maximum surfaces wind 143 speed is greater than 62 km/h then it is considered as a severe thunderstorm event otherwise if wind speeds are 144 between 31 and 62 km/h then it is considered as an ordinary thunderstorm case (also used by Saha et al. (2014)). 145 Hereafter, the total number of thunderstorm occurrences per year in both severe and ordinary category is 146 counted and represented by thunderstorm frequencies as TSS and TSO. Here it may be noted that, the wind 147 speed measurements are taken from the first measurement of radiosonde balloon flight for all stations. These 148 datasets are always within 10m from the surface and according to WMO criterion, they can assume a maximum 149 error of 1 m/s from surface to 100 hPa level. Since a minimum wind speed of 31kmph or 8.61 m/s is required 150 for identification as an ordinary thunderstorm, hence this 1 m/s error is not expected to perturb the thunderstorm 151 severity climatology presented in this study. 152 IMD provides daily rainfall accumulations in 0.25 degree spatial resolution over the Indian region since 153 the year 1900 (Rajeevan et al., 2006, 2008; Pai et al., 2014). This daily precipitation data at the closest grid point 154 is used to define the frequency of severe and weak rainfall days hereafter referred to as WRF and SRF 155 respectively. The severe rainfall frequencies constitute those days where the daily accumulation is greater than

156 124.5 mm/day while for the weak rainfall cases it is less than 7.5 mm/day according to IMD glossary as given in 157 http://imd.gov.in/section/nhac/termglossary.pdf . 158 From the previous section it follows that a set of 14 parcel parameters with rainfall and thunderstorm 159 frequencies are essential to understand the convective climatology over India. However, other than this, 8 160 standard instability parameters (LI, KI, TTI, CT, VT, CAPE, CINE and MLC) are also additionally important to 161 quantify the thunderstorm severity, hence must also be considered for analysis. Now, it is known that most of 162 these instability parameters are inter-related; hence principal component analysis (PCA) analysis is done to 163 identify and use only those instability parameters that can give a complete but independent overview of the 164 atmospheric instability using minimum parameters. In this analysis, introduced by (Hoteling 1936) a set of 165 possibly related parameters are converted into orthogonal independent components after which the primary 166 components are plotted with the initial parameters. Parameter variance scores present at the farthest distance 167 from the primary principal components and also from all the other variables contain the highest variance; hence 168 they are selected for representing the existing group of old inputs. Hence in the present study, daily datasets of 169 all 6 instability parameters are averaged to yearly values for every regions and then the PCA analysis is 170 performed on the datasets. Daily datasets have not been directly used for PCA as it would have too many 171 fluctuations which would make the redundant parameter identification very difficult in all cases. The variance 172 distribution plot (not shown) for each region showed that only the first two components contribute to more than 173 70% of the total variance; hence the covariance scores of these two strongest orthogonal components are plotted 174 in Fig. 2 which depict that the LI is completely unrelated to rest of the parameters. Again, since VT is found to 175 lie exactly in the middle of the rest of the parameters, and it also represents the lower tropospheric instability in 176 a much more suitable way hence this parameter is also used with LI to represent the rest of the instability 177 parameters in a convenient way.. Consequently, LI and VT are additionally considered along with the previous 178 set of 14 attributes to get the final set of 16 parameters for further analysis. 179 Thus, a set of 16 parameters are finally taken for the analysis: LCL, LFC, EL, LI, VT, CAPE, CINE, 180 MLC, PWV, PWL, WSH, T100, TSO, TSS, WRF and SRF. However, apart from the IGRA radiosonde and the 181 IMD rainfall database, it was believed that some other parameters may also be externally responsible for the 182 changing trends in atmospheric instability and hence they are also included. They comprise the monthly mean 183 aerosol absorption index (AAI) data taken from the Tropospheric Emission Monitoring Internet Service 184 (TEMIS) Air Pollution Archive (De Graaf et al., 2005). In addition, the monthly average gridded data of ozone 185 mixing ratio (OMR) and Specific Humidity (SHUM) along with Downward Long Wave Radiation Flux 186 (DLWRF) are also utilized from ERA-Interim Re-analysis datasets 187 (https://apps.ecmwf.int/datasets/data/interim-full-daily/levtype=sfc/). 188 We have estimated all parameters from daily radiosonde data and averaged over a season and annually 189 for obtaining trend at 95% confidence interval using robust regression analysis (Shepard, 1968). Further, the 190 parameters from radiosonde were averaged region wise and then the robust fit algorithm is employed on the 191 normalized time series to get the long-term trends (Andersen, 2008; Raj et al., 2018). These yearly trend values 192 are multiplied by 37 to get the total climatological trend in one parameter over the complete data span of 1980- 193 2016. For seasonal trend analysis, the same approach has been utilized for different seasons. The seasonal 194 distribution has been adopted from IMD reports which are as follows: Pre-monsoon (March-May), Monsoon 195 (June-September), Post-monsoon (October-November) and Winter (December-February). Further, for studying

196 the periodicities associated with each of these time series, an Empirical Mode Decomposition (EMD) technique 197 is used (Wu and Huang, 2009). Finally, the robust fit analysis is done on each of these components to compare 198 the trends from each periodicity to determine which of the periodicities dominates in each parameter. 199 200 3. Results 201 3.1. Climatic trends over Chennai 202 In the previous study by Chakraborty et al. (2018), long term trends of instability were investigated over 203 Gadanki (13.5oN,79.2oE) situated on a hilly terrain with an altitude of 370 m above sea level at a distance of 204 ~150 km from the eastern coasts and Bay of Bengal. To see whether, the observed trends of these parameters are 205 behaving similarly in case of IGRA profiles also the climatic trends of instability are now described over 206 Chennai (13.08oN, 80.27oE) which is the closest radiosonde station from Gadanki. The yearly averaged datasets 207 are normalized with respect to their climatic mean and are plotted along with 1 sigma standard errors in Fig 3 208 after which robust fit regression analysis (Andersen 2008) is utilized to obtain the climatological trends in these 209 parameters as shown by red solid lines in the plots. A decreasing trend in VT and increase of magnitude in 210 CINE with LI is noticed which indicates a reduction in the lower atmospheric instability (Fig.3d,e,h,i). 211 However, CAPE (Fig.3f) shows significant increasing trends throughout the period. LFC has a slightly 212 ascending trend (~18 hPa) which leads to increasing CINE and decreasing VT over Chennai, while the EL is 213 found to get lifted up drastically (Fig.3c) resulting in an increase in the total instability and CAPE. The increase 214 in height of EL can be caused by a reduction in temperatures in the upper tropospheric heights (Manohar et al, 215 1999). Hence, it can be inferred that the reduction in temperatures near 100 hPa (Fig.3l) plays an important role 216 in modulating the total atmospheric instability and CAPE. 217 The enhancement in CINE magnitude and reduction in VT leads to the reduction in the frequency of 218 weaker convective systems with medium or lower CAPE values. Again, as CAPE is one of most important 219 parameters that modulate convective severity, hence the frequency of severe thunderstorms and heavy rainfall 220 occurrences is expected to rise (Fig.3n,p). Thus, it is inferred that lower level instability has reduced due to 221 elevated CINE and LFC; while the upper-level atmospheric instability has intensified significantly due to a 222 cooling at 100 hPa and ascension in EL over Chennai. Hence, CAPE value increases drastically leading to more 223 severe thunderstorm and heavy rainfall frequency events during the mentioned period. 224 Before proceeding to the investigation on the climatological trends of convection and instability over 225 the Indian region, it is necessary to validate whether the obtained hypothetical trends from Chennai are free 226 from any data quality issues. Hence a region wise climatology of the most important parameter CAPE is 227 obtained from all the Indian regions using ERA-Interim Reanalysis data and the trends are shown in Fig. S1. 228 This figure clarifies that all the Indian regions (especially the coastal regions) have experienced a common rise 229 in CAPE especially after 1996-2000. Thus, the stated hypothesis looks clear and hence this can be progressed 230 over a much broader way. 231 However, it should be noted that Fig 3 provides too much detailed and cumbersome results related to 232 all 16 parameters and the complexity of the analysis is expected to increase further when similar analysis will be 233 presented for all the Indian regions together. On the other hand, for a complete understanding about the 234 morphology of upper and lower tropospheric instability, all the instability parameters will be required. Hence, to 235 reduce chances of confusion and to make the results more compact, all 16 parameters will be discussed together

236 but only a few of them will be presented in the main study. After a thorough consideration with respect to the 237 main objective of the present attempt, 8 parameters namely LFC, EL, CAPE, CINE, PWV, T100, TSS and SRF 238 are retained in the main figures while their complementary aspects such as LCL, LI, VT, MLC, PWL, WSH, 239 TSO and WRF are shown in the supplementary sections. 240 3.2. Climatological average of parameters 241 The climatological mean values of all instability parameters over six different Indian sub-divisions are 242 shown with boxplot analysis (McGill, 1978) in Fig. 4 and Fig. S3. The utility of using this approach is that, it 243 will reveal which regions of India shows normal expected variation (if it lies within the box limits signifying 25- 244 75% percentage of the distribution), while on the other hand it will also identify those regions having the 245 extreme outlier values (lying outside the whiskers signifying the outermost 5% of the distribution). The LCL 246 (Fig.S3a) and LFC (Fig.4a) are found to be at the lowest altitudes in the coastal regions. As these stations 247 receive most of the moisture from Sea, the EL (Fig.4b) is also expected to be higher at the coastal areas and 248 lower elsewhere. However, due to low moisture availability, the inland regions experience weaker instability 249 which results in lower CAPE (~900 J/kg) (Fig.4c) with higher CINE (Fig.4d) and WSH (Fig.S3f). During strong 250 convection, the values of LI (Fig.S3b) (which represents that the mid-tropospheric instability) are also expected 251 to be more negative in the coastal regions. Similarly, height integrals of instability such as CAPE (Fig.4c) and 252 MLC (Fig.S3d) are significantly higher (~1500 J/kg) in the coastal regions while the magnitude of MLC 253 (Fig.S3d) is found to be almost half of CAPE. As the trends in total convective strengths below 300 hPa are 254 quite low compared to that over the total atmospheric column, hence it follows that the portion of buoyant 255 column above 300 hPa must have contributed significantly to the total convective developments over the Indian 256 region. Again, being opposite of CAPE, CINE values are minimum in the coastal regions compared to inland 257 and continental regions thereby serving as a potential cause for the reduced instability in these regions. 258 Similar to CAPE and MLC combination, the PWV (Fig.4e) and PWL (Fig.S3e) pair shows the highest 259 averages in the coastal regions due to their closest proximity to the adjoining seas. Also, PWL (moisture integral 260 up to 700 hPa) is found to be almost half of PWV, hence the mid and upper tropospheric humidity is found to 261 play a strong role in modulating the convective systems over India. The instability and moisture are highest in 262 the coastal regions hence the frequency of severe thunderstorms and rainfall occurrences are comparatively 263 higher (Fig.4g,h). The North Western region shows the large values of thunderstorm frequency which is not 264 supported by other parameters. Hence, it may be inferred that this is due to frequent dry storms called “Andhi” 265 which have no relation with convective instability and rainfall (Rajpal and Deka, 1980). Thus, it can be 266 concluded that the effect of convection is large in the coastal regions compared to other regions which resulted 267 in high CAPE with more thunderstorms and intense rain occurrences. 268 3.3. Long-term trends in the instability parameters 269 The long-term trends are calculated for each parameter during the entire study period of 1980-2016 for all 270 regions using the robust regression analysis at 95% confidence interval as depicted in Fig. 5 and Fig. S4. For 271 simplicity, the average trends along with their standard deviation values are depicted in Table 1. Also to 272 investigate about the significance of trend values calculated from these time series datasets, a t-tset analysis 273 (Gosset, 1908) is done on all parameter and locations. The p values are calculated at 95% confidence limits for 274 t-test analysis on all instability parameters over the Indian sub-divisions and interestingly, all the values are 275 found to be below 0.05. Hence the time series variations to be presented in subsequent sections will always be

276 statistically significant in nature. So, to have a better quantitative measure of the trend significance, the total 277 changes in each of these parameters are presented in percentage form in place of the p values in the table. This 278 process will enable an easy identification of regions experiencing more accelerated convective growth. But on 279 the other hand, while analyzing the results of the trend analysis in statistical form, the absolute trend has to be 280 given more importance as the % changes completely depend on the magnitude of the long term mean. The LCL 281 (Fig.S4a) height is found to decrease which may lead to an overall increase in the number of rain occurrences 282 throughout the country (provided that the amount of atmospheric instability is adequate). On contrary, LFC 283 (Fig.5a) is found to ascend in all the regions except NE resulting in the reduction of lower level instability and 284 an increase of CINE magnitude (Fig.5d). However, the extent of change in LFC (Fig.5a) and LCL (Fig.S4a) is 285 smallest in the coastal regions (~10 hPa). In case of EL (Fig.5b), a very prominent ascent is depicted in all 286 regions (highest in coastal regions) which increase the height of the buoyant column; hence the net effect on 287 total instability and CAPE (Fig.5c) is expected to increase significantly. Similarly, LI (Fig.S4b) values become 288 more negative in all the regions with slightly higher magnitudes in the coastal regions. VT represents the lower 289 level atmospheric instability and hence is expected to be affected by the elevation in LFC. Thus, a reduction in 290 VT (Fig.S4c) is seen with minimum values in the coastal regions (~0.3), medium in the NE and NW regions 291 (~0.5) and highest in deep inland regions such as CI and PI (~0.8). An intensification in CAPE (Fig.5c) is seen 292 in all regions (~1100 J/kg) as expected from EL (Fig.5b) and LI. However, the increase is the highest (~100%) 293 at the coastal regions whereas in MLC (Fig.S4d), which is measured only up to 300 hPa level, the increment is 294 only 20% of that in CAPE. Hence, it follows that the maximum contribution towards the increase in CAPE 295 comes above 300 hPa. In case of CINE, an overall enhancement in values is observed as expected (~60 J/kg). In 296 addition, the trend values suggest a two-fold increase of CINE in inland regions while the values are much lesser 297 (50%) in the coastal regions due to balancing effect from strong convections and CAPE in those regions. 298 The PWV (Fig.5e) and PWL (Fig.S4e) values are increasing similar to CAPE and MLC. The long-term 299 trends in PWV are about 10% of its climatological average with highest in the coastal regions. Further, the 300 lower level moisture content of PWL (up to 700 hPa) showed an increase but the trend values are comparatively 301 smaller (~6%). As it has been made clear that it is not the lower tropospheric moisture (below 700 hPa) but the 302 remaining amount which is increasing significantly at par with CAPE for all regions, hence there may be a 303 possible association between these two factors which needs to be investigated in the coming sections. The WSH 304 (Fig.S4f) parameter increases in all regions of the country, and hence it produces an inhibiting effect on the 305 lower level instability. An upper tropospheric cooling trend is observed in all other regions (Fig. 5f) with 306 minimum values in the inland regions and maximum in the coastal regions. Consequently, the increase in CAPE 307 values is maximum in the coastal regions and lesser elsewhere. The ordinary thunderstorm frequency is also 308 found to increase (Fig. S4g) which may be due to the partial damping effect of an elevated LFC and CINE on 309 lower level instabilities. However, the TSS (Fig.5g) is found to increase at a much higher rate compared to TSO 310 especially in the coastal regions. On the other hand, an increase in CINE and decrease of VT lead to an increase 311 in the number of WRF (Fig.S4h). However, due to rise in CAPE and TSS (Fig.5g), the SRF (Fig.5h) is also 312 found to rise significantly by about 20% particularly in the coastal regions. It may be noted that, as EL has more 313 dominant effect on CAPE hence the rise in SRF is much larger than that WRF (5%). Finally, the long-term 314 trends have been compared between the east and west coastal regions and it is observed that the rate of increase 315 in total instability is the most prominent in the western coasts while factors related to ascending LFC, CINE and

316 reducing VT are more significant in the central India which is the farthest from both the sea coasts. Thus, the 317 long-term analysis infers that lower atmospheric instability has reduced while the upper tropospheric instability 318 and moisture increased drastically over the Indian region. As a result, convective severity as expressed in terms 319 of CAPE, TSS and SRF is increasing more strongly in the coastal regions while in the continental areas this 320 effect is dampened due to the contribution of increasing CINE and WSH. 321 3.4. Seasonal effect on long-term trends in the instability parameters 322 The seasonal variation of the long-term variations of atmospheric instability is shown in Fig. 6 and Fig. 323 S5. LCL shows a uniform descent by 10 hPa in all seasons (Fig.S5a) whereas LFC ascends in most of the 324 regions and seasons (Fig.6a). However, this ascent is more prominent in the monsoon and post-monsoon season. 325 However, the seasonal variation is absent in EL and LI (Fig.6b, Fig S5b) which are mainly associated with an 326 upper layer phenomenon. VT shows the most prominent reduction in monsoon and post-monsoon seasons 327 (Fig.S5c). MLC and CAPE show a lot of regional disparities but with a common increase in its value in all the 328 seasons (Fig.6c, S5d). In monsoon and post-monsoon, the increase in CAPE is slightly lesser due to the effect of 329 decreasing VT and elevated LFC. CINE is closely related to VT and LFC, hence it shows slight increase (of 330 magnitude) in the monsoon and post-monsoon seasons with maximum values in inland regions as expected 331 (Fig.6d). 332 PWV, PWL and WSH represent a prominent increase in the monsoon followed by the post-monsoon 333 (Fig.6e, Fig.S5e-f). T100 is related to an upper atmospheric phenomenon hence no seasonal or spatial variation 334 is displayed, except for a small cooling effect in pre-monsoon (Fig.6f) due to the prevalence of intense 335 convections events which is supported by the strongest increase in CAPE. A decrease in lower atmospheric 336 instability and increase in CINE is observed; hence TSO and WRF are expected to increase. However, this 337 increase is found more dominant only in the monsoon and post-monsoon (Fig.S5g,h). Another interesting result 338 is that TSS and SRF do not behave similarly. TSS increases almost uniformly in all seasons with the highest in 339 the pre-monsoon. However, SRF increases mainly in the monsoon followed by the post-monsoon season 340 (Fig.6g,h). The observed disparity between them is due to the profuse moisture availability during monsoon and 341 post-monsoon compared to the pre-monsoon. 342 Further, in seasonal trends, east and west coasts show equivalent trends in all instability parameters 343 while the Central India still remains as the region which is most affected by the ascension of LFC and CINE. 344 Thus, the seasonal analysis reveals that the yearly long-term trends are almost uniformly distributed in all the 345 seasons. The ordinary and weak thunderstorm frequencies show the strongest increase during monsoon and 346 post-monsoon while the upper atmospheric instability shows a weak influence in the pre-monsoonal trends on 347 the yearly climatology. 348 3.5. Effect of specific periodicities on long-term trends 349 In case of both annual and seasonal trend analysis, all Indian sub-divisions are found to follow similar 350 behavior. Hence, to find out the periodicities in the average long-term trends, the time series of all regions are 351 averaged and then subjected to EMD technique which reveals the existence of four main periodicities namely: 352 1.5 - 2.5 years corresponding to QBO, 4-6 years corresponding to ENSO, 10-12 years corresponding to the solar 353 cycle and the fourth one is of 16-20 years. A similar multi-decadal climatic oscillation was also reported by 354 Dhaka et al. (2010). Hence for simplicity, this periodicity has been renamed as a Multi-decadal Climatic 355 Oscillation (MCO).

356 The climatic trends of these periodicities for each parameter are calculated from robust regression 357 analysis. An illustration of the obtained MCO periodicities for CAPE along all the Indian regions is shown in 358 Fig. S2. Further for comparison, the trend values from each periodicity is normalized to percentage with respect 359 to the total trend values for each parameters and the net contribution of these individual periodicities are 360 depicted in Fig. 7 and Fig. S6. The figure suggests that ENSO, QBO and solar cycle have no effect on LCL 361 (Fig.S6a) while MCO is quite strong. LFC shows minimal effects to all periodicities except solar cycle period 362 which may be due to solar-terrestrial heating (Fig.7a). EL and LI are significantly affected by both solar and 363 MCO periodicities (Fig.7b, Fig, S6b). But in LI, the contribution from MCO is much more than solar effect. In 364 case of VT (Fig. S6c) the effect of both ENSO and MCO are found prominent. CAPE is found to be strongly 365 influenced by MCO followed by solar effect (Fig.7c) and this is also discernible from the most strong increasing 366 trends in CAPE especially in the coastal regions after the years 1996-2000 in Fig. S2. However, in case of MLC, 367 contribution of MCO is comparatively lesser (Fig.S6d) hence some separate phenomena above 300 hPa may 368 have prominent influence on increasing CAPE. Apart from CAPE, effect of MCO is also found very strong in 369 case of CINE (Fig.7d). 370 The moisture parameters like PWV and PWL show similar variability as in CAPE and MLC which 371 indicates significant moisture transport changes only above 300 hPa in the past 18 years (Fig.7e, Fig S6e). WSH 372 does not show the dominance of any periodicity (Fig.S6f) while T100 shows the most prominent contribution 373 from the MCO (Fig.7f) thereby showing its connection with the long-term variability in EL and CAPE with 374 associated thunderstorm severity in the recent years. TSO and TSS are both affected by solar and MCO 375 (Fig.S6g, Fig.7g) but TSS shows that the effect of MCO is higher compared to TSO. Finally, the effect of MCO 376 is also found more prominent in case of SRF and WRF (Fig.7h, Fig S6h). In nutshell, the MCO acts as the most 377 dominant periodicity which has influenced the convective severity over India. So, in the coming sections, the 378 MCO trends for both halves of 37 years will be studied, For ease of indication and referencing, these trends of 379 18 years span each will be hereafter mentioned as quasi-bi-decadal trends (since both spans are close to 20 years 380 in length). 381 382 3.6. Investigation of quasi-bi-decadal trends between 1980-1997 and 1999-2016 383 In the previous section, the annual averaged time series of many parameters such as EL, LI, VT, CAPE, 384 CINE, T100, TSS, WRF and SRF showed very significant changes with respect to MCO. It has also been 385 indicated from Fig. S2 that the climatic trends before and after the period 1996-2000 are significantly different 386 from each other. Therefore, the trends have been estimated with respect to two time periods before and after the 387 year 1998. The time series for both MCO are produced and their trend values are represented in Fig. 8 and Fig. 388 S7. For simplicity, the MCO are referred as C1 (1980 to 1998) and C2 (1999 to 2016), respectively. Starting 389 with LCL, in C1 there is almost no change, but in C2 there is a strong descent which influences the overall 390 change in the time series (Fig.S7a). In case of LFC, C1 shows an ascending trend, but in C2, a significant 391 increasing pattern of LFC pressure is seen hence an overall descent is obtained (Fig.8a). An ascent in the EL is 392 noticed in both the periods however during C2 the trends show significant enhancement (Fig.8b). LI values are 393 expected to become more negative from 37 years trend, however its absolute magnitude shows a slight reduction 394 in C1 followed by a prominent increase in C2 resulting in a net increase in instability (Fig.S7b). VT shows an 395 overall decreasing pattern in both the periods (Fig.S7c). CAPE (Fig.8c) shows an enhancement in both the

396 cycles but the trends become more prominent in C2 (1500 J/Kg). Similar to CAPE, MLC (Fig.S7d) also shows 397 an increasing trend in both the cycles but the trend values are also much smaller than in CAPE. Hence, the rise 398 in EL height can be considered as a primary factor for increase in CAPE above 300 hPa during C2. CINE shows 399 increasing trend in both C1 and C2 but again the trend values are much stronger (~80 J/kg) during C2 especially 400 in the inland regions (Fig.8d).. 401 The moisture trends in both PWV and PWL have shown a constant increase in both the MCO 402 throughout India (Fig.8e, Fig S7e). The WSH (Fig.8k) also increases uniformly in both MCO with strongest 403 trends in the inland regions. A prominent cooling of ~1.5 degrees is seen in 100 hPa levels everywhere in C1, 404 but in C2 the trend increases to ~-2.5 degrees (Fig.8f) which can be considered responsible for the abrupt 405 elevation in EL and increasing CAPE values during the recent years. TSO increases slightly in C2 compared to 406 C1 (Fig.S7g). But in case of TSS, the positive trend gets doubled in C2 mainly in the coastal regions (Fig.8g). 407 Finally, in case of SRF the trend values in C2 are slightly higher with the maximum magnitudes in the coastal 408 regions as expected (Fig.8h). A further comparison between the six regions reveals that the west coast shows the 409 maximum enhancement in all the instability and convective severity parameters in the past 18 years due to 410 strong growth in moisture content and associated cooling at 100 hPa. 411 On the contrary, during C2 central India suffers from the maximum reduction in lower level instability 412 as seen from the rise in CINE and LFC due to the dearth of moisture content. Similar results are also found in 413 other coastal and inland regions. Hence it follows that mainly during C2, the upper tropospheric instability has 414 enhanced everywhere while the lower tropospheric instability has reduced which has led to the development in 415 both CAPE and CINE. As a result both TSS-TSO and WRF -SRF combination increases. 416 417 4. Discussion

418 From the previous section, it is inferred that a cooling trend at 100 hPa levels has led to the ascent in EL 419 which results in an increase in CAPE, TSS and SRF. To explain the reason behind this, we consider the ozone to 420 be a primary heating agent by absorbing the incoming solar ultraviolet radiation near 100 hPa level 421 (Mohanakumar, 2008). OH hydroxyl radicals are formed by oxidation of water vapor molecules by a reactive 422 oxygen atom at the same height. On the other hand, it has been reported by Forster et al. (2007) that in the recent 423 years there has been a cooling in upper troposphere due to decrease in ozone concentration near 70 hPa. Hence, 424 it is hypothesized that the OH radicals formed from the oxidation of water vapour can take an active role in the 425 breakup of ozone molecules at 100 hPa levels which may lead to this cooling effect. The preceding sections 426 have shown a significant increase in moisture content especially in the coastal areas hinting towards more 427 moisture transport from the adjoining seas. Again, an increase in LI and CAPE values have also been reported in 428 most of the regions which can lift the available moisture to upper atmospheric levels (Das et al., 2016; Guha et 429 al., 2017). To add to this increasing CAPE and LI, many recent researchers’ have reported a net increase in the 430 Hadley cell and Brewer-Dobson circulation strength (Liu et al., 2012; Fu et al., 2015; Shepherd and 431 McLandress, 2011) which also assists in the up-liftment of moisture to upper atmospheric levels. Thus, it is 432 inferred that low-level moisture is transported to the upper troposphere and above where it is responsible for 433 ozone depletion and cooling thereby elevating the EL and increasing the thunderstorm severity. 434 To test this hypothesis, yearly averaged time series of specific humidity and ozone mixing ratio data are 435 collected for all stations and the quasi-bi-decadal trend values are depicted in Fig. 9. This figure shows a rise in

436 specific humidity levels by 7% in C2 over entire India (Fig.9a). On the other hand, trends of specific humidity 437 have almost trebled in C2 phase with the maximum values in the coastal regions (Fig.9e). As water vapor 438 concentration increases, ozone concentration is expected to decrease. The ozone trends support this hypothesis 439 by showing a sharp transition from low positive to high negative values during C2 (Fig.9b,f). It may be 440 additionally noted that the specific humidity increase and reduction in ozone content are strongest in the coastal 441 regions leading to a higher increase in CAPE and severe thunderstorms in those regions. 442 In the recent decades, Indian region has experienced a surface warming trend which is mainly caused 443 by an increase in greenhouse gas concentrations as pointed out by Basha et al. (2017). These greenhouse gases 444 are heat absorbing in nature and these particles reside within the lower troposphere (generally below 700 hPa) 445 due to surface heating and boundary layer dynamics as reported by Chakraborty et al. (2017b). Further, these 446 gases has a tendency to absorb and then re-emit the outgoing longwave radiation as emitted by the Earth 447 resulting in more downward longwave radiation flux and atmospheric heating which elevates the LFC. 448 Additionally, this near surface heating reduces the vertical temperature lapse rate leading to a drop in lower 449 instability (VT). To test this hypothesis, yearly averaged Downward Long Wave Radiation Flux (DLWRF) time 450 series is depicted over the Indian region in Fig. 9(c,g) which also suggests that DLWRF values are increasing in 451 C2. To show that the increase in DLWRF is due to the heat absorbing particles only, the trends in Absorbing 452 Aerosol Index (AAI) are shown for all the regions. The figure suggests that the mean of AAI is increasing 453 slightly more in C2 with a positive trend (Fig.9d,h). Due to this heating of lower atmosphere and capping of 454 lapse rates by greenhouse gases and absorptive aerosols, the LFC starts ascending, so WSH and CINE get 455 stronger while VT reduces. As a result the ordinary to weak convective occurrences start increasing. 456 Finally, it has to be explained why the upper air instability and CAPE are increasing mainly in the 457 coastal regions. The coastal regions have high moisture content (Saha et al., 2017). Because of the strong land- 458 ocean contrast, low-level winds close to 850 hPa flow into the coastal regions and disperse the pollutants and 459 greenhouse gases to other locations leading to a weaker convective inhibition in those areas. This hypothesis is 460 supported by the lowest AAI values in the coasts despite having high increasing trends in those areas. In 461 addition, the ample moisture supply in the coastal regions is lifted up to the upper troposphere and lower 462 stratosphere (UTLS) where it undergoes prominent cooling due to ozone reduction. Hence, the EL ascends more 463 resulting in higher CAPE which finally led to an abrupt rise in TSS and SRF in the coastal regions. However, in 464 the inland regions the layer of absorptive aerosols and greenhouse gases cannot be dispersed amply due to the 465 dearth of strong lower level winds. As a result, the growth of lower atmospheric instability gets inhibited in the 466 inland regions. Further, due to less moisture availability, UTLS cooling and EL ascent are much lower hence 467 there is a less rise in CAPE which ultimately leads to an increase in TSO and WRF in those sub-divisions. It 468 may be noted that the trend in AAI is not signiﬁcantly different for the two time periods C1 and C2. Again, it is 469 the EL and not the LFC or LCL which influences CAPE strongly; hence the strong trends of humidity increase 470 and ozone reduction overpowers the weaker inhibitory effect from the atmospheric aerosols and this acts as a 471 major driving force behind the increase in convective severity compared to in most of the cases. 472 473 5. Summary and conclusions

474 In recent decades, global warming has become a threat to human life and society in terms of its various 475 implications. Increase in surface temperature leads to stronger atmospheric instabilities which in turn may

476 increase the CAPE resulting in more severe thunderstorm and precipitation activity. Hence, the long term 477 variations of instability parameters will help to better understand the changes in the weather extremes with 478 respect to climate change. In light of the above, the main objective of the present study is to analyze whether 479 convective instability is changing over the Indian region during the last 37 years, and then to find its possible 480 effects on thunderstorm and rainfall severity. Radiosonde measurements from Integrated Global Radiosonde 481 Archives (IGRA) pertaining to 27 stations across 6 Indian sub-divisions are utilized to depict the spatial 482 distribution of these long-term trends during the period of 1980-2016. The selection of instability parameters is 483 done based on Principle Component Test (PCT) which showed the importance of taking LI and VT for further 484 investigations. A total of 16 parameters (including parcel and instability data with moisture content, wind shear, 485 and thunderstorm and rainfall frequencies) have been utilized. Robust fit approach is employed on the regional 486 average time series to calculate the long-term trends on both yearly and seasonal basis. The main highlights 487 obtained from the present study are listed below: 488 1. The coastal regions experience the most significant rise in Convective Available Potential Energy (CAPE) 489 and Equilibrium Level (EL) leading to more occurrences of Severe Thunderstorms (TSS) and severe 490 rainfall events (SRF) while the inland regions undergo a decrease in lower atmospheric instability due to 491 elevated Convective Inhibition Energy (CINE) and Level of Free Convection resulting in more Ordinary 492 thunderstorm (TSO) and Weak Rainfall occurrences (WRF). 493 2. In the pre-monsoon season, an increasing TSS activity is observed due to higher instability connected to 494 increasing EL height and CAPE values, along with a decrease in LI values while, the monsoon and 495 postmonsoon season experiences more prominent ascension in LFC height with larger values of CINE, 496 Wind Shear (WSH) thereby increasing the tropospheric stability which lead to increased TSO and WRF 497 occurrences all over the Indian region. 498 3. The Empirical Mode Decomposition (EMD) analysis on the instability parameters reveals that the 16-20 499 year multi-decadal oscillation (MCO) as the most dominant component in all six Indian sub-divisions. 500 4. The quasi-bi-decadal analysis reveals an increase in magnitude in many parameters like EL, CAPE, CINE, 501 TSO and TSS along with cooling at 100 hPa level during C2 (1999-2016) which dominates 37-year trend. 502 5. The annual and quasi-bi-decadal trends support that the increase in thunderstorm severity and associated 503 convection is strongest along western coasts due to maximum moisture ingress from the seas while the 504 greatest reduction in lower atmospheric instability is experienced in central India owing to the lack of 505 pollutant dispersal as it is situated very far from the seas. 506 6. In the coastal regions, ample amount of water vapor is advected into the mid-troposphere from the 507 surrounding seas which in presence of strong lifting goes up to upper troposphere and lower stratosphere 508 (UTLS) where ozone depletion occurs leading to a strong cooling effect. This cooling effect enables the 509 ascent in EL resulting in much stronger LI and CAPE values, hence more TSS and SRF.. 510 7. In the inland regions, the dispersing effect by sea winds is absent hence the capping effect of lower 511 instability is more leading to stronger CINE values. Again, due to the dearth of moisture transport from the 512 seas, the UTLS cooling is lesser; leading to a weaker rise in CAPE consequently the TSO and WRF 513 frequencies increase significantly. 514 8. However, as the ascent in EL has a stronger contribution over increasing CAPE than the inhibitory effect 515 of LFC, hence the long term trends are expected to be more strongly influenced by the ozone

516 decomposition and cooling at 100 hPa levels than the capping effect of low level inversions from 517 absorptive aerosols; hence the convective severity over the Indian regions is found to increase. 518 Thus, it may be inferred that in the near future also, convective severity will increase strongly in the 519 coastal regions while weak and ordinary thunderstorms will be more common in the inland regions. It may 520 appear at certain sections of this analysis that the trends of CAPE and EL are exorbitantly high; but it is not the 521 actual case because previous studies by Murugavel et al (2012) and Gettlemann et al. (2002) have also shown 522 almost comparable trends in convective severity both in India and abroad. Nevertheless, this study gets an upper 523 hand over the previous approaches as it successfully explains the hypothesis brought forward by early research 524 attempts that UTLS cooling at 100 hPa and greenhouse gases concentration rise can regulate the climatic trends 525 of convective severity and frequency especially over tropical regions in the recent decades. 526 After going through the study, there may be a possibility of thinking that the change in instability 527 trends is due to the change in sensors around 1998. But this is not the actual case because first, there has not 528 been any mention in past literature survey related to any change in radiosonde data quality during late 1990s in 529 IMD or IGRA. Secondly, the yearly variations of all 16 parameters for various IGRA stations as in Chennai do 530 not commonly show any abrupt change in time series during 1996-2004 except for a few cases. Thirdly it has 531 been revealed by IMD reports that the year 2000 was a tipping point for the climate change led warming over 532 India thereby leading to a rise in catastrophic weather events and a cataclysmic fallout will follow by the year 533 2040 if these emission scenarios are not curbed recently (Hindustan Times, 2019). Thus, it follows that the 534 observed changes in the atmospheric instability trends before and after 1996-2004 are due to a synoptic global 535 warming based climate change phenomena and not due to any change in radiosonde sensor type. 536 However, in spite of all this, the present study has certain shortcomings. The most important one 537 among them is that this set of explanations is based on isolated information from selected in-situ observations 538 and hence it needs to be studied in more detail spatially in future using model-based observations. In the recent 539 years, certain studies have utilized multiple GCM outputs over the US to infer the robust increase in 540 thunderstorm frequency (Diffenbaugh et al., 2013; Seeley and Romps, 2015). However, these types of studies 541 have not yet been done over the Indian region. Hence, a combination of multi-station radiosonde data with 542 model data will be utilized to provide a generalized picture about convective severity over the Indian region. On 543 the other hand, this study also introduces the effect of direct aerosol heating on instability and convection; but 544 the probable impact of indirect aerosol loading on modulating the cloud lifetime and convective severity has not 545 been discussed here. This is because, the relationship between indirect aerosol forcing and instability is still very 546 unclear and complex (Connoly et al. 2012). A few researches in the recent years have hypothesized that a higher 547 concentration of aerosols may lead to stronger updrafts velocities by altering the latent heat release resulting in 548 growth of CAPE and TSS (Tao et al. 2012; Storer and van den Heever, 2013). However, this is a season and 549 location specific phenomena and hence it is not expected to impact the yearly trend of CAPE and TSS as strong 550 as the upper tropospheric cooling effect projected in this study. But in future, an exhaustive analysis of cloud 551 and aerosol components involving both in-situ and modelled data can to be done to investigate its contribution 552 on the total CAPE, TSS and SRF trends over the Indian region. 553

554 Author contributions: Rohit Chakraborty had analysed complete data and written first draft, Ghouse Basha 555 helped in the analysis and corrections and Venkat Ratnam supervised overall the work including final 556 corrections. 557 558 Acknowledgments 559 One of the authors (Rohit Chakraborty) thanks, Science and Engineering Research Board, Department 560 of Science and Technology for providing fellowship under National Post-Doctoral Scheme (File 561 No:PDF/2016/001939). He also acknowledges National Atmospheric Research Laboratory, for providing 562 necessary support and data for this work. The authors also thank S.T. Akhil Raj, Sanjeev Dwivedi and N. 563 Narendra Reddy for their suggestions. Data used in present study can be obtained directly from IGRA website. 564 565 References

566 Alappattu, D.P. and Kunhikrishnan, P.K.: Premonsoon estimates of convective available potential energy over 567 oceanic region surrounding Indian subcontinent. J. Geophys. Res., Atmos. 114, 2009.

568 Ananthakrishnan, R.: Some aspects of the monsoon circulation and monsoon rainfall. Pure Appl. Geophys., 569 115(5–6), 1209–1249, 1977.

570 Andersen, R.: Modern methods for robust regression. 152, Quantitative Applications in Social Sciences, Sage, 571 2008.

572 Basha, G., Kishore, P., Ratnam, M. V., Jayaraman, A., Kouchak, A. A., Ouarda, T. B. M. J., and Velicogna, I.: 573 Historical and Projected Surface Temperature over India during 20 th and 21 st century. Sci. Rep., 7(1), 574 2987, 2017.

575 Brooks, H. E.: Severe thunderstorms and climate change. Atmos. Res., 123, 129–138, 2013.

576 Chakraborty, R. and Maitra, A.: Retrieval of atmospheric properties with radiometric measurements using 577 neural network. Atmos. Res., 181:124–132, https://doi.org/10.1016/j.atmosres.2016.05.011, 2016.

578 Chakraborty, R., Talukdar, S., Saha, U., Jana, S., and Maitra, A.: Anomalies in relative humidity profile in the 579 boundary layer during convective rain. Atmos. Res., 191, https://doi.org/10.1016/j.atmosres.2017.03.011, 580 2017a.

581 Chakraborty, R., Saha, U., Singh, A. K., and Maitra, A.: Association of atmospheric pollution and instability 582 indices: A detailed investigation over an Indian urban metropolis. Atmos. Res., 583 https://doi.org/10.1016/j.atmosres.2017.04.033, 2017b.

584 Chakraborty, R., Basha, G., Ratnam and M.V.: Diurnal and long-term variation of instability indices over a 585 tropical region in India. Atmos. Res., https://doi.org/10.1016/j.atmosres.2018.03.012, 2018.

586 Connolly PJ, Vaughan G,May PT, Chemel C, Allen G, Choularton TW, Gallagher MW, Bower KN, Crosier J, 587 Dearden C. 2013. Can aerosols influence deep tropical convection? Aerosol indirect effects in the Hector 588 island thunderstorm. Q. J. R. Meteorol. Soc. 139: 2190–2208. DOI:10.1002/qj.2083.

589 Das, S.S., Ratnam, M.V., Uma, K.N., Patra, A.K., Subrahmanyam, K.V., Girach, I.A., Suneeth, K.V., Kumar 590 K.K., and Ramkumar, G.: Stratospheric intrusion in to the troposphere during the tropical cyclone Nilam 591 (2012), Q. J. R. Meteorolog. Soc. doi: 10.1002/qj.2810, 2016.

592 Dhaka, S. K., Sapra, R., Panwar, V., Goel, A., Bhatnagar, R., Kaur, M.,: Influence of large-scale variations in 593 convective available potential energy (CAPE) and solar cycle over temperature in the tropopause region at

594 Delhi (28.3oN, 77.1oE), Kolkata (22.3o N, 88.2o E), Cochin (10o N, 77o E), and Trivandrum (8.5oN, 77.0oE) 595 using radiosonde during 1980--2005. Earth Planets Space, 62(3), 319–331, 2010.

596 De Graaf, M., Stammes, P., Torres, O., and Koelemeijer, R. B. A.: Absorbing Aerosol Index: Sensitivity 597 analysis, application to GOME and comparison with TOMS. J. Geophys. Res. Atmos., 110(D1), 2005.

598 Diffenbaugh, N. S., Scherer, M., and Trapp, R. J.: Robust increases in severe thunderstorm environments in 599 response to greenhouse forcing. Proc. Natl. Acad. Sci., 110(41), 16361–16366, 2013.

600 Durre, I., Vose, R. S., and Wuertz, D. B.: Overview of integrated global radiosonde archive. J. Clim., 19(1), 53– 601 68, 2006.

602 Emanuel, K.: Climate and tropical cyclone activity: A new model downscaling approach. J. Clim., 19(19), 603 4797–4802, 2006.

604 Ferreira, A.P., Nieto, R., Gimeno, L., Completeness of radiosonde humidity observations based on the IGRA, 605 Earth Syst. Sci. Data Discuss,., 2018, https://doi.org/10.5194/essd-2018-95.

606 Forster, P. M., Bodeker, G., Schofield, R., Solomon, S. and D. Thompson: Effects of ozone cooling in the 607 tropical lower stratosphere and upper troposphere, Geophys. Res. Lett., 34, L23813, 608 doi:10.1029/2007GL031994, 2007.

609 Fu, Q., Lin, P., Solomon, S., and Hartmann, D. L.: Observational evidence of the strengthening of the Brewer- 610 Dobson circulation since 1980. J. Geophys. Res. Atmos., 120(19), 2015.

611 Gosset, W.S., The probable error of a mean, Biometrika, 6(1), 1-25, 1908.

612 Gensini, V. A., and Mote, T. L.: Downscaled estimates of late 21st century severe weather from CCSM3. Clim. 613 Change, 129 (1-2), 307-321, 2015.

614 Gettelman, A., Seidel, D. J., Wheeler, M. C., and Ross, R. J.: Multidecadal trends in tropical convective 615 available potential energy. J. Geophys. Res. Atmos. 107(D21), 2002.

616 Hindustan Times: https://www.hindustantimes.com/environment/freak-weather-to-rise-in-india- 617 over-two-decades/story-T1G8SgfBh8jydT15UnKGuM.html, last access: 24 January, 2019.

618 Hotelling, H., Analysis of complex statistical variables into principal components, J. Educational Psychology, 619 24, 417-411, 1936.

620 Guha, B. K., Chakraborty, R., Saha, U., and Maitra, A.: Tropopause height characteristics associated with ozone 621 and stratospheric moistening during intense convective activity over Indian sub-continent. Global Planet. 622 Change, 158. https://doi.org/10.1016/j.gloplacha.2017.09.009, 2017.

623 Huntrieser , H., Schiesser, H.H., Schmid, W.,and Waldvogel, A.: Comparison of Traditional and Newly 624 Developed Thunderstorm Indices for Switzerland, Weather Forecasting, 12, 108-125, 1997.

625 Joseph, P. V., Raipal, D. K., and Deka, S. N.: Andhi. The Convective Dust Storms of Northwest India. Mausam, 626 31, 431–442, 1980.

627 Kharin, V. V, Zwiers, F. W., Zhang, X., and Wehner, M.: Changes in temperature and precipitation extremes in 628 the CMIP5 ensemble. Clim. Change, 119(2), 345–357, 2013.

629 Liu, J., Song, M., Hu, Y., and Ren, X.: Changes in the strength and width of the Hadley circulation since 1871. 630 The climate of the past, 8(4), 1169–1175, 2012.

631 Manohar, G. K., Kandalgaonkar, S. S., and Tinmaker, M. I. R: Thunderstorm activity India and Indian 632 southwest monsoon. J. Geophys. Res. Atmos. 104(D4), 4169–4188, 1999.

633 McGill, R,Turkey, J.W., Larsen, W.A., Variations of Box Plots, The American Staistician. 32(1), 12-16, 1978.

634 Mohanakumar, K.:Stratosphere-troposphere interactions: introduction.Springer Science Business Media, 2008.

635 Murthy, B.S. and Sivaramakrishnan, S.: Moist convective instability over the Arabian Sea during the Asian 636 summer monsoon, 2002, Meteorol. Appl., https://doi.org/10.1017/S135048270500201X, 2006.

637 Murugavel, P., Pawar, S. D., and Gopalakrishnan, V.: Trends of convective available potential energy over the 638 Indian region and its effect on rainfall. Int. J. Climatol., 32(9), 1362–1372, 2012.

639 Nelli, N. R., Ratnam, M.V., Basha, G., and Ravikiran, V.: Cloud vertical structure over a tropical station 640 obtained using long-term high resolution radiosonde measurements, Atmos. Chem. Phys. Discuss., 641 https://doi.org/10.5194/acp-2018-194, 2018a.

642 Nelli, N. R., and Rao, K.G..: Contrasting variations in the surface layer structure between the convective and 643 non-convective periods in the summer monsoon season for Bangalore location during PRWONAM, J. 644 Atmos. Sol. Terr. Phys., https://doi.org/10.1016/j.jastp.2017.11.017, 2018b.

645 Pai, D. S., Sridhar, L., Rajeevan, M., Sreejith, O. P., Satbhai, N. S., and Mukhopadhyay, B.: Development of a 646 new high spatial resolution (0.25 x 0.25) long period (1901--2010) daily gridded rainfall data set over India 647 and its comparison with existing data sets over the region. Mausam, 65(1), 1–18, 2014.

648 Prein, A. F., Rasmussen, R. M., Ikeda, K., Liu, C., Clark, M. P., and Holland, G. J.: The future intensification of 649 hourly precipitation extremes. Nat. Clim. Change, 7(1), 48, 2017.

650 Raj, S. T. A., Ratnam, M. V., Rao, D. N., and Murthy, B. V. K.: Long-term trends in stratospheric ozone, 651 temperature, and water vapor over the Indian region. Ann. Geophys. (Vol. 36, p. 149), 2018.

652 Rajeevan, M., Bhate, J., Kale, J. D., and Lal, B.: High resolution daily gridded rainfall data for the Indian 653 region: Analysis of break and active monsoon spells. Current Science, 296–306, 2006.

654 Rajeevan, M., Bhate, J., and Jaswal, A. K.: Analysis of variability and trends of extreme rainfall events over 655 India using 104 years of gridded daily rainfall data. Geophys. Res. Lett., 35(18), 2008.

656 Rao, Y. P.:Southwest monsoon. Synoptic Meteorology, 1976.

657 Riemann-Campe, K., Fraedrich, K., and Lunkeit, F.: A global climatology of convective available potential 658 energy (CAPE) and convective inhibition (CIN) in ERA-40 reanalysis. Atmos. Res., 93(1), 534–545, 2009.

659 Saha, U., Maitra, A., Midya, S. K., and Das, G. K.: Association of thunderstorm frequency with rainfall 660 occurrences over an Indian urban metropolis. Atmos. Res., 138, 240–252, 2014.

661 Saha, U., Chakraborty, R., Maitra, A., and Singh, A. K.: East-west coastal asymmetry in the summertime near- 662 surface wind speed and its projected change in future climate over the Indian region. Glob. Planet. Change, 663 152., https://doi.org/10.1016/j.gloplacha.2017.03.001, 2017.

664 Santhi, Y. D., Ratnam, M. V., Dhaka, S. K., and Rao, S. V.: Global morphology of convection indices observed 665 using COSMIC GPS RO satellite measurements. Atmos. Res., 137, 205–215, 2014.

666 Seeley, J. T., and Romps, D. M.: The effect of global warming on severe thunderstorms in the United States. J. 667 Clim., 28(6), 2443–2458, 2015.

668 Shepard, D.: A two-dimensional interpolation function for irregularly-spaced data. In Proceedings of the 1968 669 23rd ACM national conference (pp. 517–524), 1968.

670 Shepherd, T. G., and McLandress, C.: A robust mechanism for the strengthening of the Brewer-Dobson 671 circulation in response to climate change: Critical-layer control of subtropical wave breaking. J. Atmos. Sci., 672 68(4), 784–797, 2011.

673 Storer, R.L., van den Heever, S.C.: Microphysical processes evident in aerosol forcing of tropical deep 674 convective clouds. J. Atmos. Sci., 70, 430–446, 2013 doi.org/10.1175/JAS-D-12-076.1

675 Tao, W.K., Chen, J.P., Li, Z., Wang, C., Zhang, C., 2012. Impact of aerosols on convective clouds and 676 precipitation. Review of Geophysics, 50, RG2001. doi:10.1029/2011RG000369

677 Wang, S.-Y., Davies, R. E., Huang, W.-R., and Gillies, R. R.: Pakistan’s two-stage monsoon and links with the 678 recent climate change. J. Geophys. Res. Atmos., 116(D16), 2011.

679 Webster, P. J., Holland, G. J., Curry, J. A., and Chang, H.-R.: Changes in tropical cyclone number, duration, and 680 intensity in a warming environment. Science, 309(5742), 1844–1846, 2005.

681 Wu, Z., and Huang, N. E.: Ensemble empirical mode decomposition: a noise-assisted data analysis method. 682 Advances in Adaptive Data Analysis, 1(1), 1–41, 2009.

683 Xie, B., Zhang, Q., and Ying, Y.: Trends in precipitable water and relative humidity in China: 1979--2005. J. 684 Appl. Meteorol. Clim., 50(10), 1985–1994, 2011.

685

686

Name CI EC NE NW PI WC India µ ơ % µ ơ % µ ơ % µ ơ % µ ơ % µ ơ % µ ơ % LCL (hPa) 39 2.8 4.43 8.9 0.1 0.93 15.4 0 1.71 24 0.6 2.63 45.5 1.0 5.17 9.1 0.2 0.96 23 6.4 2.49 LFC (hPa) -38 1.4 5.57 -11 0.2 1.41 13.4 0.1 1.91 -44 3.8 6.63 -9.2 0.0 1.36 -17 0.6 2.19 -18 8.7 2.46 EL (hPa) -188 11.2 49.2 -280 27.1 82.8 -206 8.8 60.5 -230 2.0 67.6 -239 6 63.6 -311 40 91.4 -242 19 68.1 LI (oC) -0.8 0.01 16.7 -1.7 0.2 22.2 -1.3 0.1 22.8 -1.1 0.1 17.7 -1.4 0.0 27.2 -1.8 0.1 24.6 -1.3 0.1 20.3 VT (oC) -0.7 0.01 2.98 -0.3 0.02 1.50 -0.5 0 2.24 -0.5 0.0 2.32 -0.9 0.1 3.85 -0.4 0.0 1.95 -0.5 0.1 2.34 CAPE (J/kg) 617 2.9 82.8 1589 90.8 108 1137 30 125 858 53 90.3 1000 39 107 1554 198 98.9 1126 159 97.9 MLC (J/kg) 55 0.24 12.1 288 9.6 42.8 273 4.8 56.7 134 1.1 29.7 201 16 43.7 323 27 42.0 212 42 36.8 CINE (J/kg) -94 7.4 87.8 -36 0.3 46.7 -30 1.2 27.8 -85 6.9 103 -67 2.4 62.6 -44 1.5 55.3 -59 11 73.7 PWV (mm) 1.4 0.03 5.71 3.2 0.03 10.0 3.7 0.1 13.4 1.3 0.0 4.72 2.2 0.0 8.97 3.9 0.1 11.2 2.6 0.5 8.85 PWL (mm) -0.2 0 1.95 0.4 0 2.75 0.7 0.1 6.22 0.0 0 4.65 0.6 0.0 5.71 0.6 0.0 3.98 0.4 0.2 3.32 WSH (/s) 5.8 0.3 78.4 3.4 0.2 54.4 4.8 0.2 75.0 3.4 0.0 54.4 5.5 0.6 74.8 3.6 0.1 68.6 4.4 0.4 69.8 T100 (oC) -1.5 0.03 3.00 -2.5 0.1 5.20 -0.4 0 0.83 -0.3 0 0.59 -2.5 0.3 5.13 -2.2 0.2 4.68 -1.6 0.4 3.00 TSO 1.4 0.05 24.3 0.8 0 10.5 2.2 0.0 53.1 3.5 0.2 53.8 2.7 0.2 40.9 0.5 0 4.92 1.8 0.5 27.3 TSS 1.5 0.0 70.5 2.3 0.05 144 2 0.1 250 2.5 0.3 209 1.7 0.1 81.7 2.3 0.1 131 2.1 0.2 147 WRF 2.9 0.1 9.51 3.8 0.1 6.55 4.6 0.4 11.5 2.2 0.0 4.19 2.8 0.1 8.88 6.3 0.2 11.4 3.8 0.6 7.28 SRF 0.4 0.0 32.4 0.8 0.06 22.2 0.2 0.0 8.30 0.2 0 14.8 0.2 0 14.4 1.1 0.1 39.5 0.5 0.2 20.5 687

688 Table 1: Statistical information related to the 37-year trend of all instability parameters over the six sub-divisions of India (µ: long-term average, ơ: standard 689 deviation, %: total percentage trend).

Figures

Figure 1: The locations of the 27 IGRA stations used for the present study. The distribution of the 27 stations over Indian regions is as follows: 4 stations in the NC, 6 stations in EC, 4 stations in NE, 4 stations in the NW, 5 stations in the PI and finally 4 stations in WC.

CI EC NE (a) CAPE (b) (c) MLC 0.5 0.5 CAPE 0.5 CAPE MLC LI MLC CT CT LI TTI CT TTI 0 0 TTI 0 KI VT LI CINE KI VT CINE KI VT -0.5 -0.5 CINE -0.5 Principal Component 2 -0.5 0 0.5 -0.5 0 0.5 -0.5 0 0.5 NW PI WC (d) (e) (f) CAPE MLC MLC CAPE MLC 0.5 CT 0.5 CT 0.5 CAPE TTI LI TTI CT KI TTI 0 VT 0 KI LI 0 KI LI CINE VT VT CINE CINE -0.5 -0.5 -0.5 Principal Component 2 -0.5 0 0.5 -0.5 0 0.5 -0.5 0 0.5 Principal Component 1 Principal Component 1 Principle Component 1

Figure 2: Principle Component Analysis for selection of instability parameters for the long-term trend study in (a) Central India (CI), (b) East Coast (EC), (c), North East (NE), (d) North West (NW), (e) Peninsular India (PI) and (f) West Coasts (WC) obtained using IGRA observations.

Figure 3: Long-term variation of (a) LCL, (b) LFC, (c) EL, (d) LI, (e) VT, (f) CAPE, (g) MLC, (h) CINE, (i) PWV, (j) PWL, (k) WSH, (l) T100, (m) TSO, (n) TSS, (o) WRF and (p) SRF observed over Chennai.

1 2 Figure 4: Climatological mean values of (a) LFC, (b) EL, (c) CAPE, (d) CINE, (e) PWV, (f) T100, (g) TSS and 3 (h) SRF over the six sub-divisions of India. Coastal Regions are represented by red cones, the north eastern 4 and western regions are denoted by black stars and diamonds while the blue cones represent the inland 5 regions. Here the box limits refer to the upper and lower quartiles (25% and 75%) while the whiskers refer to 6 the outlier limit of the data (5% and 95% limit of the population) 7

Figure 5: Long-term variation of (a) LFC, (b) EL, (c) CAPE, (d) CINE, (e) PWV, (f) T100, (g) TSS and (h) SRF over the six sub-divisions of India during the period 1980-2016. Legends are same as in Figure 4.

Figure 6: Seasonal trend of long-term variation for (a) LFC, (b) EL, (c) CAPE, (d) CINE, (e) PWV, (f) T100, (g)

TSS, and (h) SRF over India during all seasons. Here 1 refers to pre-monsoon (March-May), 2 refers to Monsoon

(June-September), 3 for Post-monsoon (October-November) and 4 for Winter (December-February). Legends are

same as in Figure 4.

Figure 7: Percentage contribution of various periodicities on long-term trend of all instability parameters over India

namely: 1.5 -2.5 years periodicity denoted as 1, 4 -6 years periodicity denoted as 2, 10-12 years periodicity

displayed as 3 and 16-20 years periodicity represented as 4.

Figure 8: Comparison of average values for two time periods indicating the trend of various instability parameter

over the six sub-divisions of India in two half periods of 18 years each (the numbers 1 and 2 represent the ﬁrst and

second period, C1 and C2, during 1980-1997 and 1999-2016, respectively) during 1999-2016. Legends are same

as in Figure 4.

Figure 9: Average values and climatological trends of specific humidity, ozone mixing ratio at 100 hPa and Downward Longwave Radiation Flux (DLWRF) with Absorptive Aerosol Index (AAI) over the six sub-divisions of India over two half periods of 18 years each (1980-1997 and 1999-2016). Legends are same as in Figure 4.

Appendix

Table A1: Details of the dataset used.

Altitude Initial No. of No. of Sl Latitu Longit above No. of profiles profile Station Station Name Region no. de ude MSL profiles at 00 & s at available 12Z 00Z 1 42361 26.23 78.25 205 Gwalior 9901 9412 4530 2 42369 26.75 80.88 122 Lucknow 16869 16387 8963 Central 3 42379 26.75 83.37 78 Gorakhpur 12376 11793 6170 India 4 42667 23.28 77.35 522 Bhopal 14795 13968 4472 5 42809 22.65 88.45 6 Kolkata 15212 14626 6980 6 42971 20.25 85.83 45 Bhubaneshwar 18325 17552 6672 70 Vishakhapatna 7 43150 17.68 83.3 13225 12856 6355 m Eastern Coasts 8 43185 16.2 81.15 3 Machilipatnam 17108 16374 8014 9 43279 13 80.18 14 Chennai 14067 13487 8278 10 43346 10.92 79.83 7 Karaikal 16519 16106 6890 11 42314 27.48 95.02 110 Dibrugarh 10067 9550 3801 12 42410 26.1 91.58 54 Guwahati 15280 14803 8681 North 13 42492 25.6 85.17 51 Patna 8934 8318 4370 Eastern 14 42724 23.88 91.25 16 Agartala 15234 14732 6340 15 42101 30.33 76.47 251 Patiala 11572 10129 4663 16 42182 28.58 77.2 210 New 14077 13982 6581 North 17 42339 26.3 73.02 217 Jodhpur 13133 12918 5274 Western 18 42647 23.06 72.63 55 Ahmadabad 11430 11006 5540 19 42867 21.1 79.05 310 Sonegaon 15626 14971 8532 20 43014 19.85 75.4 585 Aurangabad 14220 13993 4032 Peninsular 21 43041 19.08 82.03 554 Jagdalpur 10568 10205 5437 India 22 43128 17.45 78.47 530 Hyderabad 10234 9723 6195 23 43295 12.97 77.58 917 Bangalore 10150 9514 4899 24 43003 19.12 72.85 14 Bombay 14102 13808 7030 25 43192 15.48 73.82 58 Goa 7070 6313 5180 Western 26 43285 12.95 74.83 31 Mangalore 9866 9406 5020 Coasts 27 43371 8.48 76.95 60 Trivandrum 11590 11120 8304

Table A2: List of Abbreviations

Slno. Abbreviation Full Form 1 LCL (hPa) Lifted Condensation Level 2 LFC (hPa) Level of Free Condensation 3 EL (hPa) Equilibrium Level 3 LI (oC) Lifted Index 4 VT (oC) Vertical Totals Index 5 CAPE (J/kg) Convective Available Potential Energy 6 MLC (J/kg) Mixed Layer CAPE 7 CINE (J/kg) Convective Inhibition 8 PWV (mm) Precipitable Water Vapour 9 PWL (mm) Lower Level PWV 10 WSH (s-1) Wind Shear 11 T100 (oC) Temperature at 100 hPa 12 TSO Ordinary Thunderstorm Frequency 13 TSS Severe Thunderstorm Frequency 14 WRF Weak Rainfall Frequency 15 SRF Severe Rainfall Frequency 16 SHUM (kg/kg) Specific humidity 17 AAI Absorptive Aerosol Index 18 IMD India Meteorological Department 19 IGRA Integrated Global Radiosonde Archive 20 GHG Green House Gas 21 DLWRF (W/m2) Downward Long Wave Radiation Flux 22 EMD Empirical Mode Decomposition 23 UTLS Upper Troposphere Lower Stratosphere 24 QBO Quasi-biennial oscillation 25 ENSO El Niño–Southern Oscillation